20.3 Abstracted versus Claims-Based Quality Data

20.3 Abstracted versus Claims-Based Quality Data

Quick Answer: Chart-abstracted measures come from trained reviewers applying a specification to the record. Claims-based measures come from ICD-10-CM/PCS codes, present-on-admission (POA) indicators, and other claim fields. Inpatient CDI moves claims-based results strongly because the codes are the data. Agency for Healthcare Research and Quality (AHRQ) Patient Safety Indicators (PSIs) are claims-based administrative algorithms, not abstracted nursing process measures.

Mortality indices (Sections 20.1–20.2) already showed one claims-risk pathway: expected deaths rise or fall with coded severity. Domain VIII also tests whether you can tell how a quality number was born. Hospitals run abstracted core measures, electronic clinical quality measures, infection-surveillance rates, and claims algorithms in the same committee. If you treat every rate as “something quality nurses pulled from the chart,” you will mis-state how CDI can—and cannot—move the number. Chapter 21 covers which CMS programs use which measures. This section covers the data source, which is the skill you need before those program names mean anything.

Two pipelines (and a third electronic stream)

Chart abstraction means a trained abstractor (often in quality, registry, or a specialty program) reads the medical record—and sometimes laboratory, radiology, or device data—against a written specification. Inclusion, exclusion, numerator events, and timing come from that specification, not from whether a diagnosis code appears on the UB-04 / 837I claim. Historical Joint Commission and CMS chart-abstracted core measures (acute myocardial infarction, heart failure, pneumonia, surgical care) worked this way. Many have been retired or shifted toward electronic clinical quality measures (eCQMs), which pull structured electronic health record data rather than a human worksheet. eCQMs are not “claims” and not classic abstraction; they are a third stream. CDI still helps when the record is ambiguous, but the eCQM engine is looking for discrete fields, timestamps, and coded EHR elements specified by the measure—not for an MCC.

Claims-based measures run software on administrative data. Typical inputs:

• ICD-10-CM diagnosis codes and ICD-10-PCS procedure codes
• POA indicators on principal and secondary diagnoses
• Age, sex, discharge status, admission source, length of stay
• Sometimes prior claims for comorbidities (CMS 30-day models)

If the diagnosis was never coded, or POA is wrong, the algorithm never sees the clinical truth. That is why concurrent and retrospective CDI, POA integrity, and clinical validation sit at the center of claims-based quality work.

Why CDI hits claims-based measures so hard

A claims algorithm is a recipe. Ingredients are codes and claim flags. CDI and coding control those ingredients for the inpatient stay.

Missing specificity leaves the recipe without a numerator event—or without a risk factor that would have excluded the case or raised expected risk. Example: postoperative acute respiratory failure never named; a respiratory PSI or mortality model never receives the diagnosis the ventilator flowsheet described.

Wrong POA flips in-hospital versus present-on-admission logic. A pressure injury that was present at the inpatient order but billed POA = N can look like a hospital-acquired event in claims software. A true in-hospital event billed POA = Y can vanish from a hospital-acquired claims measure. For the CMS hospital-acquired condition (HAC) payment provision, N and U generally prevent the condition from acting as a CC/MCC; W is paid like Y. That payment rule is not identical to every PSI exclusion, but it shows why POA is not a clerical afterthought (Chapter 16 and Chapter 17).

Unsupported codes create false numerator events. Clinical validation of a coded complication is quality integrity, not “anti-quality.” A PSI or HAC flag built on a diagnosis the record cannot support is an audit and ethics problem even if the dashboard looks worse after you remove it.

Risk adjustment on the claim is the mortality-index lesson again. CMS 30-day mortality and readmission models, and many vendor tools, use diagnosis codes from the index stay and sometimes from prior claims. CDI that captures true comorbidities can change expected outcomes on those models the same way ROM capture changes in-hospital O/E.

Queries still follow the 2026 standard: no quality-outcome language. “Document this to drop a PSI” is noncompliant. “The chest radiograph on hospital day 2 shows a new pneumothorax after central-line placement; please clarify whether an iatrogenic pneumothorax was diagnosed, another explanation applies, or the finding is still uncertain” is a clinical clarification. The PSI software will do what the eventual codes and POA tell it to do.

PSIs are claims-based—do not recast them as abstracted nursing measures

AHRQ Patient Safety Indicators are software specifications that use administrative discharge data. Current CMS Hospital-Acquired Condition Reduction Program composite PSI 90 uses PSIs 03, 06, 08–15 (handbook shorthand: PSIs 03, 06, and 08–15; live CMS pages list PSI 08 as in-hospital fall-associated fracture). Chapter 21 names components. What Domain VIII needs here is the source: those indicators are not collected by bedside nurses ticking a PSI flowsheet, and they are not the same thing as National Healthcare Safety Network (NHSN) healthcare-associated infection rates.

Confusion is understandable in real hospitals. After the PSI software flags a case, quality staff often open the chart to see whether the event was real, preventable, or a coding problem. That post-flag review is good operations. It does not convert the PSI specification into a chart-abstracted core measure. The official numerator was still triggered by codes and POA (and related claim fields). If CDI and coding never put the triggering diagnosis on the claim, there is no flag for the abstractor to second-guess.

NHSN central line-associated bloodstream infection (CLABSI), catheter-associated urinary tract infection (CAUTI), selected surgical site infection (SSI), methicillin-resistant Staphylococcus aureus (MRSA) bacteremia, and Clostridioides difficile infection (CDI) used in the HAC Reduction Program are surveillance rates. Infection prevention applies Centers for Disease Control and Prevention (CDC) definitions. Coding a complication code is neither necessary nor sufficient for an NHSN event. Conversely, an NHSN event can exist when coding used a different, still-correct diagnosis. Do not teach “the CAUTI PSI is what the wound nurse abstracts.” (There is not a one-to-one “CAUTI PSI equals NHSN CAUTI” identity in any case; AHRQ and NHSN use different recipes.)

The HAC Reduction Program mixes claims-based PSI 90 with NHSN infection measures. The CMS HAC payment provision (fourteen categories, POA-driven grouping effect) is a third construct. Teach them as separate machines that sometimes discuss the same clinical topics (pressure injury, falls, infections). Merging them into “HACs, which nurses abstract” fails Domain VIII and Domain VI at once.

What CDI should do in each pipeline

For claims-based work: concurrent review of POA, specificity of complications versus present-on-admission disease, linkage of organ failure, procedure-diagnosis relationships the algorithm uses, and clinical validation of codes that create flags. For abstracted work: make the record internally consistent so an abstractor can apply the spec—times, contraindications, reasons a measure does not apply—without relying on a leading query. For NHSN: do not “code the infection into existence”; do support accurate clinical documentation of infections that were treated, and leave case determination to the surveillance definition.

Hybrid CMS measures that combine claims with EHR data exist. They do not erase the basic split: if the published spec is claims-driven, CDI’s coded-data work is in the critical path; if the spec is chart- or EHR-driven, coding the MS-DRG will not by itself complete the measure.

Traps

• Stating that PSIs are abstracted nursing quality indicators.
• Stating that NHSN HAIs are assigned by the MS-DRG or APR-DRG grouper.
• Assuming a post-flag chart review means the PSI specification is chart-abstracted.
• Believing CDI cannot affect claims-based measures because “quality owns the dashboard.”
• Believing CDI should write “PSI exclusion” into the query.
• Treating every quality rate as claims-based, including classic abstracted core measures and eCQMs.
• Inventing unpublished PSI numerical thresholds or implying a single national O/E cut score for PSI 90.